Presupposition and Policing in Complex Demonstratives
Forthcoming in Nous
Michael Glanzberg (Univeristy of Toronto) and Susanna Siegel (Harvard)
January, 2004
The expressions this cat and that glove with a hole are complex demonstratives.
In this paper, we defend a thesis about complex demonstratives. The thesis we defend
concerns the role of the nominal (e.g., cat and glove with a hole) in a central class of uses.
In the utterances at issue, we argue, the nominal F in that F plays a policing role: no
proposition is semantically expressed by the utterance, if the object appropriated by the
speaker’s use of that F fails to be F. We'll call this nominal policing.1[1]
In characterizing nominal policing, we introduced the concept of an appropriated
object. Roughly, a speaker’s use of that F appropriates an object, if the speaker
demonstrates that object, where demonstrations can include both publicly observable
gestures by the speaker as well as speaker intentions. For example, if a speaker says That
car is better than that car, successively pointing at and intending to talk about two cars in
plain view, with each use of that car the speaker appropriates a different object. In this
case, the object is a car both times. If in place of one of the cars, there was a boat, but the
speaker’s intentions and gestures of pointing remained the same, the speaker would
appropriate a boat by one of her uses of that car. Nominal policing predicts that the
speaker’s utterance in this case would fail to semantically express a proposition.
The thesis of nominal policing has been a focal point of debate within the
literature on complex demonstratives. It has been defended by David Braun (1994) and
Emma Borg (2000), while Richard Larson and Gabriel Segal (1995) have suggested that
it is false.2[2] Much of this debate has relied upon the view that complex demonstratives,
like bare demonstratives, are referring expressions, and paradigms of direct reference at
that. If such a view is assumed, then at least part of the semantic contribution of a use of
a complex demonstrative is its referent, and the crucial question becomes what, if any,
semantic contribution the complex demonstrative can make beyond its referent. Policing,
or the denial of policing, become theses about the status of this additional contribution.3[3]
Now, the nominal in complex demonstratives can play a policing role, even if
complex demonstratives are not directly referential. Recently, philosophers have begun to
consider seriously the view that complex demonstratives are quantificational, where this
1[1]
We put linguistic items in italics, and their corresponding semantic values in upright
text.
2[2]
Braun’s view is a bit more subtle than we suggest: we say that the utterance fails to
semantically express a proposition due to the nominal failing to be satisfied, whereas
Braun would say that it expresses a gappy proposition, which is in turn either false or
truth-valueless.
3[3]
The locus classicus for direct reference is David Kaplan’s “Demonstratives”
(1977/1989). There he says relatively little about complex demonstratives. As Braun
(1994) remarks, the view of “Demonstratives” appears to imply policing. Braun’s own
paper offers a proposal compatible with Kaplan’s that does include policing (cf. note 2).
is held to exclude their being devices of reference (direct or otherwise).4[4] Our defense
of nominal policing, unlike previous ones, is neutral on whether uses of complex
demonstratives are referring expressions (as direct reference theorists hold),
quantificational expressions (as King has proposed), or discourse anaphors in a dynamic
semantics (as Craige Roberts (2002) has argued). This neutrality is not a lack of interest,
and its point is not simply to avoid a premise which some have recently denied. Rather,
we think that understanding what underlies nominal policing is fundamental to
understanding the behavior of complex demonstratives, and in particular, the behavior
that makes them appear referential. We will suggest that this behavior itself can be
understood without supposing that complex demonstratives have the semantics of
referring expressions.
Our defense of nominal policing will focus on a certain sort of presupposition
failure to which uses of complex demonstratives are prone, when the appropriated object
fails to satisfy the nominal. We argue on independent grounds that presupposition failure
of this sort is tantamount to failure to semantically express a proposition. This will be
central to our defense of nominal policing.
The discussion will proceed as follows. After some preliminary remarks in
Section 1, in Section 2 we consider and reject the view that the nominal in complex
demonstratives plays no truth-conditional role whatsoever, and some other related views.
In Section 3, we present the core evidence for nominal policing. In Section 4, we argue
that what appear to be some counterexamples to nominal policing do not really enjoy this
status. We conclude the discussion in Section 5 with a speculation: perhaps referring
expression as they have traditionally been understood do not form a genuine semantic
category.
1. 1. Preliminaries
Two things need explaining right away: the class of utterances at issue, and the notion of
appropriation that occurs in our definition of nominal policing.
First, we will focus on a range of uses of complex demonstratives: expressions of
the form that F. The nominal F can itself be complex, as in that large house or that glove
with a hole in it or that car which I saw the other day. The uses with which we will
primarily be concerned are perceptual uses: those in which the speaker (and usually the
hearer) perceives an object, upon which she thinks the truth or falsity of the utterance
depends. These are paradigmatic uses of demonstratives. We will call them “classic
perceptual uses,” and we will call the utterances in which they occur “perceptual
demonstrative utterances.”5[5]
We focus on classic perceptual uses because they exemplify what many have
taken to be the paradigmatic behavior of referring expressions. For instance, when they
occur in an utterance that has a truth-value, they are rigid. However, as we mentioned,
4[4]
This recent consideration was spurred by Jeffrey King (1999), which attacks the view
that complex demonstratives are devices of direct reference and defends the view that
they are quantificational. See also King (2001a) and Lepore and Ludwig (2000).
5[5]
Readers of Gareth Evans (1982) will recognize classic perceptual uses as a subclass of
the uses to which demonstrative expressions can be put, even disregarding the kinds of
quantificational examples presented by King (2001a).
our defense of nominal policing does not require any stance on the semantic category to
which complex demonstratives belong.
Second, our central claim is nominal policing:
If the object appropriated by the speaker’s use of that F in a perceptual
demonstrative utterance is not F, then the utterance fails to semantically
express a proposition.
‘Appropriation’ is our term for the contextual supplement required by uses of complex
demonstratives. Both bare and complex demonstratives require some sort of supplement
from the context. In the case of complex demonstratives, the nature of this supplement is
illustrated by the following contrast. Consider an utterance of:
1. 1. That key is bigger than that key.
Suppose the speaker successively points to and intends to talk about two keys in plain
view. Contrast an utterance under the same circumstances of:
2. 2. #The key is bigger than the key. 6[6]
The latter is markedly infelicitous.
The contrast just drawn between complex demonstratives and definite
descriptions suggests that the contextual supplement needed for complex demonstratives
differs from whatever sort may be needed in the case of definite descriptions.7[7] An
utterance of (2) could be infelicitous, even when the speaker successively points to and
intends to talk about two keys, both of which are already conversationally salient.8[8]
Here is a heuristic. A speaker appropriates an object o by the use of a complex
demonstrative, if she stands to o in a relation of the sort that would suffice for her to refer
to o by using a bare demonstrative. Some candidates for this relation include intending to
refer to o by the use of a demonstrative expression, and using publically accessible cues,
such as pointing, to indicate o.9[9] This is just a heuristic, as we wish to remain neutral on
what sorts of facts about the context make it the case that by her use of that F, a speaker
appropriates one object rather than another, or rather than nothing at all.
Our neutrality on the mechanism of appropriation makes us neutral on a subtle
question concerning the role that the nominal plays in appropriation. There seem to be
cases where the nominal is quite prominent in appropriation: for instance, cases in which
the speaker cannot make clear what object she wants to talk about by using a bare
We use ‘#’ to indicate markedly defective utterances. ‘?’ indicates marginally
defective utterances. We take judgments of such defectiveness by native speakers to be
on par with judgments of grammaticality as data for semantic theorizing.
7[7]
We owe this point King (2001a, Ch. 2).
8[8]
In some cases of definite descriptions as discourse anaphors, we can have a weakened
effect. Consider:
(i) ?Whenever a new mobster tries to muscle in on an old mobster’s territory, the
mobster threatens the mobster.
(This is similar to the sort of examples attributed to Hans Kamp (cf. Heim 1990).) It is
clearly preferable to say:
(ii) Whenever a new mobster tried to muscle in on an old mobster’s territory, the
OLD mobster threatens the NEW mobster.
9[9]
For discussion of what sorts of factors underlie reference in the case of bare
demonstratives, see Siegel (2002).
6[6]
demonstrative. If the mechanism of appropriation is the speaker’s intention, then
appropriation proceeds independently of the use of the nominal, and in these cases the
nominal will play a merely epistemic role, enabling addressees to discern which object
the speaker has appropriated. In contrast, suppose the mechanism of appropriation is a set
of publically accessible cues. This opens the possibility that the use of the nominal plays
a constitutive role in appropriation: a role that not merely enables the addressee to discern
which object is appropriated, but furthermore determines which object this is.
Because we are neutral on the mechanism of appropriation, we are neutral on
whether the nominal plays a constitutive or a merely epistemic role in this sort of case.
Whatever the nature is of the contextual supplement needed by complex demonstratives,
it is clear that some such supplement is needed. ‘Appropriation’ is our term for the
supplement, whatever it is. We thus content ourselves with a functional characterization
of appropriation.
Given how we have defined the class of classic perceptual uses of demonstratives,
we may expect the appropriated object in such uses to be something the speaker
perceives, and in the simplest cases, we may expect that it is common ground between
speaker and hearer which object that is. However, we wish to stress that appropriation is
fundamentally a linguistic notion, not a notion in the philosophy of mind. Appropriation
is achieved in acceptable uses of complex demonstratives. Being appropriated is a status
something can only have in a context of discourse.
Our definition of nominal policing alludes to the object appropriated by the
speaker’s use of that F. Whatever underlies appropriation, we stipulate that if any object
is appropriated by a speaker’s use of that F in a perceptual demonstrative utterance, then
a unique object is. Something analogous will hold for plural demonstratives.10[10]
So far, we have spoken as if there is such a thing as a speaker appropriating, by
her use of a complex demonstrative, an object that does not satisfy that complex
demonstrative’s nominal. (Our very definition of nominal policing assumes that this is
possible). But logical space has room for a theory of appropriation on which an object
can be appropriated by a speaker’s use of a complex demonstrative expression, only if
that object satisfies that expression’s nominal. On this theory, cases of the sort described
in the definition of nominal policing are not possible. Since we said we were neutral on
the mechanisms of appropriation, and since the theory just mentioned is a theory in part
about those mechanisms, this theory deserves comment.
Martin Davies (1982) has defended a theory about complex demonstratives
according to which there is no such thing as a speaker’s appropriating an object that is not
F by the use of that F. Like nominal policing, Davies’s view rules out that a proposition
may be expressed by an utterance of That F is G when there is no appropriated object that
is F. Indeed, the view seems to give the same predictions about truth-conditions in the
context of utterance: an utterance of That F is G will be true only if the appropriated
object is F (in the context of utterance) and G, and it will be false only if the appropriated
object is F and not G. This view, then, makes similar predictions to ours, but in a way
that makes policing appear trivial. The reasons Davies offers for his view are ones we
10[10]
We will generally restrict our attention to singular demonstratives, so we will not
formulate the condition for plural demonstratives here.
reject. 11[11] Moreover, we will present examples below which are clear cases of
appropriation in which the nominal is not satisfied.
Though we have given only a functional characterization of appropriation, we
believe this is enough to proceed to ask what role the nominal in a complex
demonstrative plays, especially when an object is appropriated.
2. Inertness and shiftiness in complex demonstratives
The notion of appropriation suggests one reason why nominal policing is contentious. If
we think of a demonstrative as a device of reference, then it may appear that
appropriation of an object suffices to secure its referent, and that there is no truthconditional role left for the nominal to play. The nominal appears to be, as we shall say,
truth-conditionally inert. Let the strong inertness thesis be the following:
A perceptual demonstrative utterance of That F is G is true in the context
of utterance if and only if the appropriated object is G.
Larson and Segal (1995, p. 213) and Neale (1993) express some sympathy for this thesis
(though they do not offer a full-fledged defense of it), and it is defended by Stephen
Schiffer (1981). According to the strong inertness thesis, no matter what the value of F
is, the truth of an utterance of That F is G depends on and only on whether the
appropriated object is G. To keep things simple, we have stated the thesis for a sentence
of the form That F is G. More generally, the strong inertness thesis holds that the
semantic contribution of an occurrence of a complex demonstrative is determined entirely
by the appropriated object, leaving no semantic role of any sort for the nominal.
In principle, the strong inertness thesis could be combined with a proviso about
appropriation to the effect that something can be appropriated by a use of a complex
demonstrative only if it satisfies the nominal (as Davies 1982 proposed). As we
mentioned earlier, this view would seem to make the same predictions as nominal
policing: when no object is appropriated, no proposition is expressed; utterances of that F
is G have truth values only if the object the speaker wants to talk about is F; they are true
if that object is G and false if it is not G.
Without this proviso concerning appropriation, the strong inertness thesis is
incompatible with nominal policing. Following our remarks at the end of Section 1, in the
rest of this paper, when we discuss the strong inertness thesis, we will assume that the
version of the thesis at issue is free of this proviso. We will thus allow that a speaker
may appropriate an object by using a complex demonstrative, when the object does not
11[11]
We take issue with a claim in the philosophy of mind made by Davies (Davies 1982,
though from recent conversations we think his current view is different). Davies claims
that a subject cannot succeed in perceiving an object without “employing a sortal
concept” in an “individuative role” (1982, p. 292). His reasons are broadly Quinean: he
thinks there is no fact of the matter about whether one is perceptually representing a
mereological sum of tomato-parts, or a time-slice of a tomato, or the front surface of a
tomato, unless one “employs” a sortal concept that the object in fact satisfies, such as the
concept ‘tomato’. Against this, we think that perceptual representations can succeed in
representing tomatoes even if the subject “employs” no sortal concept that the tomato
satisfies. (Perhaps the perceiver is under an illusion that it is a superball, or perhaps the
tomato is in a convincing baseball costume.)
satisfy the complex demonstrative’s nominal. Indeed, as we mentioned in Section 1, we
will discuss at length examples which make our assumption appear to be the only natural
one.
The strong inertness thesis takes it to be sufficient for the truth of an utterance of
That F is G that the appropriated object be G, while policing requires that it satisfy F for
the utterance to have a truth value at all. The denial of the strong inertness thesis is
compatible with nominal policing, but does not require it. One sort of semantics that
denies the strong inertness thesis allows that in a context where the object appropriated is
not F, the utterance is false. In contrast, if nominal policing holds, then such an utterance
fails to semantically express a proposition at all.12[12] All the same, in our case for
nominal policing, the first order of business is to argue against the strong inertness thesis.
The strong inertness thesis is motivated by a sort of case made familiar by
discussions of referential uses of definite descriptions (borrowing terminology from
Keith Donnellan 1966). The referential uses in question are uses of the definite
description the F where the intended referent does not satisfy F. Despite the fact that the
speaker in Donnellan’s classic examples misdescribes the object she intends to talk about,
communication proceeds unimpeded. In one of the examples, a speaker at a party says:
3. The man in the corner drinking a martini is happy.
The speaker is pointing at and intending to talk about a man who is drinking water. Yet
communication appears to be unimpeded. Thus, part of the nominal seems to be
communicatively inert. It seems to be superfluous in indicating the man about whom the
speaker intends to communicate.
In some contexts, classic perceptual uses of complex demonstratives exhibit these
same features. A fox is nosing its way through the garbage. An onlooker who mistakes it
for a badger says to someone witnessing the scene:
4. That badger is hungry.
In some such contexts, the nominal badger is not needed to enable the hearer to
understand what the speaker intends to communicate. It is not needed, for instance, if the
fox is already available to be the topic of the conversation by being mutually
acknowledged as visually prominent. 13[13] In general, if other factors already make it
12[12]
We are counting policing as a truth-conditional role for the nominal. Policing holds
that the nominal plays a non-trivial role in truth-conditional semantics, in that it plays a
semantically specified role in determining whether a proposition is expressed. Rather
than gloss this situation as one in which the nominal has no truth-conditional role, we
prefer to describe it as one in which the nominal plays a truth-conditional role (though as
we shall say later, one that is entirely presuppositional) because this role is encoded in the
truth-conditional semantics. Moreover, policing stands in contrast to the view that the
role of the nominal is purely one of aiding the hearer’s understanding of the proposition
expressed.
Although it is not built into the definition of nominal policing that policing is the
only truth-conditional role for the nominal, we think that it is. (Here we concur with
Braun 1994 and Borg 2000.)
13[13]
Example (4) is close to one given by Larson and Segal (1995, p. 213) in their
sympathetic discussion of inertness theses.
obvious to the addressee which object the speaker has appropriated in her use of that
badger, then the nominal badger is not needed to serve this purpose.
The strong inertness thesis is motivated by the idea that communicative inertness
indicates truth-conditional inertness. It takes those uses of complex demonstratives in
which the nominal is communicatively inert as the paradigm classic perceptual uses.
It is tempting to dismiss strong inertness out of hand, as one might think that a
predicate could not be semantically inert in a sentence. This idea is expressed by the
following argument:
In a compositional semantics, the semantic value of a sentence is a
function of the semantic values of its constituents. If the nominal F in a
complex demonstrative that F were inert, F could not make its usual
contribution to the truth conditions of sentences of the form, for instance,
a is F.
Call this the semantic contribution argument. Borg (2000, p. 239) makes this argument
in criticizing Larson and Segal’s support for the strong inertness thesis:
[W]e know that the parts treated as otiose [by the strong inertness thesis]
must be treated as meaningful elsewhere in our theory; that is to say, our
semantic theory independently requires rules for predicate expressions,
which will be called into play when a predicate term appears in any other
context . Yet, when appearing concatenated with a demonstrative term,
these meaning rules must simply be bypassed; despite the superficial
similarity of that F and the F, the proponent of [the semantic inertness
thesis] must hold that our theory treats only the latter as possessing a
structured meaning…14[14]
Though we deny the strong inertness thesis, we think the semantic contribution
argument against it is misguided. The argument conflates the input to semantic
composition with the output of semantic composition. This is a mistake. It is not difficult
to give a semantics of complex demonstratives which allows the nominal both to have its
usual semantic value, and yet to be inert when it is part of a complex demonstrative.
To illustrate this point, we need to consider a fragment of a semantic theory. Let
us stipulate, not implausibly, that a sentence of the form That F is G has the structure
[[[that][F]] is G]. We shall take the predicates F and G to have their usual semantic
types, as functions from objects to truth-values, and we shall assume that the only
composition rule at work is functional application. It is relatively easy to arrive at a
semantic value for that F which is not anomalous, where that has as its semantic value a
constant function on the value of F. In this case, that F will deliver the same output, no
We have modified Borg’s notation to fit our own. Josh Dever (2001) makes a
somewhat similar criticism of inertness theses. Dever assumes (against the strong
inertness thesis) that “the proposition that that F is G logically implies the proposition
that some F is G.” He then endorses what he calls a version of semantic innocence,
which he says “insists that the predicate F in that F is G make the standard F-type
contribution to content, and hence guarantees the overlap of F-ness and G-ness that
licenses the inference [from the proposition that that F is G] to the proposition that some
F is G” (p. 277, notation modified to fit our own).
14[14]
matter what value F has. This is compatible with the nominal F having the same semantic
value as it has in other constructions.
Rather than develop this theory in much more detail, let us consider what it
predicts with respect to a use of an English sentence of the form That F is G. Return to
the scene with the garbage, in which a speaker mistakes a hungry fox for a hungry
badger, and utters That badger is hungry (4). Let us first suppose that, for an occurrence
of that as part of a classic perceptual use of a complex demonstrative, there is a function
a(c) which returns the appropriated object of the context of use c. Then the semantic
value of that will be a constant function from the value of F to a(c). In notation,
[[that]]c=λF.a(c), where an expression inside double brackets superscripted with c
indicates the semantic value of that expression relative to the context c. Then, by
applying function to argument, we can compute:
5.
[[that badger is hungry]]c=1 iff
[[hungry]]c([[that badger]]c) = 1 iff
hungry(a(c))
This is true if the object appropriated by the use of that badger has the property of being
hungry.15[15]
In this fragment of a semantic theory, the nominal retains its usual semantic value.
It nonetheless satisfies the strong inertness thesis, thanks to the unusual rule for that.
Granted, the way in which the semantic value of that F depends on the semantic value of
F, in this fragment, is not very interesting. But it is compositional just the same. This
illustrates that it is possible to give a compositional semantics for That F is G with the
following three features. First, the truth or falsity of a use of That F is G does not depend
on whether that object is F; second, the semantic values of nominals need not differ from
what they are in other constructions; third, it uses only entirely standard modes of
semantic composition (indeed, only functional application). The fact that the semantic
theory sketched has these three features shows that the semantic contribution argument
does not threaten the strong inertness thesis. For all the semantic contribution argument
shows, that thesis is perfectly plausible.
Although the strong inertness thesis is not threatened by the semantic contribution
argument, it is threatened by other considerations—notably, by some pragmatic
considerations. Return once more to the fox in the garbage. Suppose that this time, the
speaker mistakes the fox for a secret agent. The fox does not look like a secret agent, in
the way in which gasoline looks like water. But the speaker has a set of idiosyncratic
beliefs: he believes that a new sort of secret agent has come to patrol the garbage, on a
hunt for potentially useful information, and that the secret agent has the capacity to make
himself look exactly like a fox.16[16] Let us suppose that the speaker in this situation is
making two sets of mistakes. First, his beliefs about the disguises of secret agents are
15[15]
We have somewhat arbitrarily chosen to sketch a semantics in which [[that]]c has
type <<e,t>,e>, taking as input a predicate and giving as output an individual. It would
be easy to give similar treatment making it a full-fledged generalized quantifier with type
<<e,t>,<<e,t>,t>>. Nothing we say is sensitive to this difference.
16[16]
So in a sense of look different from that in gasoline looks like water, the fox does
look like a secret agent: it looks to the speaker as (the speaker believes) a secret agent
would look under the circumstances that speaker believes himself to be in.
mistaken: there are no such secret agents who can disguise themselves as foxes. A fortiori
the fox in the garbage is not a secret agent. Second, let us suppose, the speaker
mistakenly believes that his addressee shares these beliefs. In fact, the addressee knows
that the creature in the garbage is a fox, and knows as well that there are no secret agents
that can disguise themselves in the way the speaker imagines.
In such a context, consider the speaker’s utterance:
6. #That secret agent is hungry.
He takes this to be both appropriate and true. More exactly, he thinks that the nominal
secret agent is an epistemically useful nominal to use, in order to communicate what he
wants to about the creature he sees in the garbage.
Once apprised of the facts, native English speakers do not tend to regard as true
the speaker’s utterance of (6) in the context described. We take this to be a datum. There
seem to be three possible explanations for it: (i) something is appropriated by the
speaker’s use of that secret agent, and the nominal plays a truth-conditional role; (ii)
something is appropriated by the speaker’s use of that secret agent, and the utterance is
true, and in judging that it is not true, speakers are confusing lack of truth with pragmatic
inappropriateness; (iii) nothing is appropriated by the speaker’s use of that secret agent.
Only explanation (ii) is compatible with the strong inertness thesis. According to
(i), since the nominal plays a truth-conditional role, the utterance is either false or truthvalue-less, and according to (iii), the utterance is truth-valueless because nothing is
appropriated. Given that only (ii) is compatible with the strong inertness thesis, the
weaker the case for explanation (ii), the weaker the reason to believe the strong inertness
thesis.
Explanation (ii) must posit a merely pragmatic role for the nominal in (6). It
cannot be part of this role to break appropriation, since by hypothesis the fox is
appropriated by the speaker’s use of that secret agent. What makes explanation (ii)
inferior to (i) and (iii) is that there is no such role for the nominal to play. There does not
seem to be any epistemic role for it to play in helping the addressee figure out which
thing is appropriated. It is easy to imagine the case in such a way that it is obvious to the
addressee that the speaker is pointing to and intending to talk about the fox. (Indeed, this
is the way we assumed the case to be.) Thus, in (6) so described, there is no further
question about which object the speaker wishes to say is hungry, and so no epistemic role
for the nominal to play.
There is a different version of explanation (ii) that might be proposed by the
proponent of the strong inertness thesis. According to this explanation, the utterance of
(6) is pragmatically odd, not because of its semantically expressed content, but because
of an implicature it generates: namely, the false implicature that o (where o is the
appropriated object) is a secret agent. This putatively implicated proposition may itself
suffer from some additional pragmatic oddity, but neither oddity nor falsehood are
features of the proposition semantically expressed by (6).17[17] We have two objections to
this view.
First, recall that one of the hallmarks of a conversational implicature is that it is
cancelable. To take, for instance, a typical scalar implicature:
7. Some of John's children are sleeping—in fact, they all are.
17[17]
A position along these lines is discussed, cautiously, by Neale (1993).
The usual implicature of the existential quantifier is canceled by the second part of the
sentence. Such explicit cancellations are never contradictory. But now consider:
8. # That secret agent is hungry, but it is not a secret agent.
This is not merely inappropriate, it is apparently contradictory.18[18] We thus do not see
the hallmark of conversational implicature in the relation between the proposition
expressed by (6) and the proposition that o is a secret agent.
Second, on the proposed view, there has to be some pragmatic mechanism that
takes as input the putatively semantically expressed (and true) proposition that o is
hungry, and delivers as output the putatively implicated proposition that o is a secret
agent. On Grice’s original formulation (1975), the mechanism is explained via the
familiar maxims of quality (do not say what you take to be false or evidentially
unsupported), quantity (do not be overly or insufficiently informative), relation (be
relevant), and manner (do not be obscure, prolix, or disorderly). But none of these
maxims is violated if an utterance of (6) simply conveys the semantically expressed
proposition. It is clear that neither quality nor quantity is violated. Likewise relation is
not (as the appropriated object is sufficiently relevant).
The only sort of implicature in the Gricean scheme for which we might find a
mechanism then seems to be a manner implicature. The maxim of manner instructs us not
to be obscure, prolix, or disorderly. It is not always easy to know if this would be
violated, but it might be suggested that the mere use of a nominal is sufficient prolixity to
trigger a manner implicature. Our objection from cancelabilty readily applies here. Like
all implicatures, manner implicatures are cancelable, but manner implicatures are
especially easy to cancel, as there is an easy, uniform means of cancellation: you just say,
‘but I did not mean to suggest anything by that, other than ...’.
Furthermore, there are striking disanalogies between the putative implicature
generated by (6) and the paradigmatic manner implicatures. In paradigmatic manner
implicatures, the manner of expression triggers the implicature, as in Grice’s example:
9. She produced a series of sounds that correspond closely with score of “The
Star-Spangled Banner.”
As Grice points out, if a reviewer chose to write (9) as opposed to “She sang ‘The StarSpangled Banner’,” the implication would be that the singer sung badly. This implicature
is thus triggered, not by the fact being reported, but rather by the manner in which it is
expressed. But not any occurrences of linguistic material suffice for manner
implicatures. Compare (9) with:
10.
10.
A. She sang the national anthem.
B. B. She sang the country's national anthem.
Only (9) triggers any sort of manner implicature. As a number of discussion of
implicature have shown, to generate a manner implicature, some sort of marked phrase is
required, where sufficiently prolixity or disorder can suffice for markedness.19[19] On the
basis of the presence of a marked phrase, speakers calculate a manner implicature
accordingly. But as we see with (14), the mere presence of linguistic material which is
18[18]
An analogous objection applies to the view that utterances of The F is G, with the F
used referentially, implicate that o is F, where o is the thing that the referentially used
definite description is used to talk about.
19[19]
See, for instance, Horn (1989) and Levinson (2000).
not strictly necessary for expressing a content is not sufficient to generate a manner
implicature. It need not be sufficiently marked. Hence, there is no prima facie reason to
think that the mere presence of a nominal triggers a manner implicature, and as our
remarks on cancellation show, there is good reason to think it does not.20[20]
In response to this pragmatic criticism of the strong inertness thesis, a fan of the
main idea behind the strong inertness thesis might try to weaken the thesis as follows:
There are some contexts in which a perceptual demonstrative utterance of
That F is G is true, if the appropriated object is G and not F.
Whereas the strong inertness thesis says it is sufficient for an utterance of That F is G to
be true, the appropriated object is G, the weakened thesis allows that there are utterances
of That F is G that are not true, even though the object appropriated by the use of that F
really is G.
We shall call this the weakened inertness thesis. As formulated, the weakened
inertness thesis is very weak: weak enough to be compatible with number of different
theses about the semantics of complex demonstratives. But the basic point of any of them
is the same. Call contexts like that of (4) (the fox/badger) reasonable: in such contexts, it
is reasonable to take the appropriated object to satisfy the nominal. Call contexts like that
of (6) (the fox/secret agent) unreasonable: in these contexts, it is unreasonable to take the
appropriated object to satisfy the nominal. The point of the weakened inertness thesis is
to have the nominal behave as if inert in reasonable contexts like (4), but not in
unreasonable contexts like (6). The nominal will thus appear to be inert in reasonable
contexts, but will play a stronger role in unreasonable ones. We will consider three ways
of implementing the weakened inertness thesis.
The first stays as close as possible to the model of the strong inertness thesis, in
that it applies the inertness semantics. But it applies the semantics selectively, taking it to
apply only in reasonable contexts. On this view, in reasonable contexts, the nominal
combines by a rule like the one discussed for strong inertness; in unreasonable contexts,
it combines by a rule that applies the nominal’s default value. So this option appeals to
context-dependent rules of composition: which composition rules apply, on this view,
depends on whether the context is reasonable or unreasonable. Rather than introduce
20[20]
Similar points can be made, mutatis mutandis, for modern re-workings of the
Gricean scheme, such as those in Horn (1989) and Levinson(2000). We believe the point
to be quite general. There is nothing uncooperative about conveying the semantically
expressed proposition, nor is there anything marked about the use of the nominal. Hence,
there is no way to calculate an implicature in any way based on the cooperative principle.
According to the strong inertness thesis, (6) means the same as That is hungry, so
the fan of the proposal under attack might insist that comparison with the bare
demonstrative, that secret agent is prolix after all. This still does not appear to be
sufficient markedness to generate an implicature. If it were, then the view would have to
predict that any use of a complex demonstrative would be marked, and generate a manner
implicature. This prediction is a non-starter in cases where it is not already obvious
(independently of her use of a complex demonstrative) which object the speaker wishes
to appropriate.
context-dependence via a rule, it makes what rule applies itself context-dependent. 21[21]
We find this highly implausible, and are inclined to dismiss it out of hand. (For any
readers not so inclined, the objection we raise against the second proposal will apply to
this one as well).
The second option avoids appealing to context-dependent composition rules.
Instead, on this option, what is sensitive to whether the context is reasonable or
unreasonable is the nominal itself, rather than any special composition rule. On this view,
in reasonable contexts, such as (4), the nominal contributes a sufficiently general property
that the appropriated object satisfies it. And in unreasonable contexts, such as (6), it
makes its normal contribution. One cost of avoiding context-dependent rules is that the
nominal is never genuinely inert, as it is according to the strong inertness thesis. For in
both kinds of context, it contributes a property, albeit a less restrictive one in reasonable
contexts. The weakened inertness thesis is weak enough that this option satisfies it.
Like the context-dependent composition rules invoked by the first option, the
context dependence invoked here is not very plausible, prima facie. Consider a perceptual
demonstrative utterance where the nominal contains the clearly context-dependent
expression tall:
11. That tall secret agent is hungry.
In the unreasonable context of (6), the nominal would contribute the usual value of tall
secret agent, where the value of tall is fixed in part by the context. Now, consider a
reasonable context for (11). There is a five-foot tall man wearing a trench coat and a hat.
By a trick of the light, he looks to be about six feet tall, and he is not in fact a secret
agent, he is a banker. The view we are considering predicts that in this context, the
nominal tall secret agent will contribute its less restrictive value. Assuming that the value
of tall secret agent is the intersection of the values of tall and of secret agent, this option
predicts that the context extends the extension of tall to five feet, and extends the
extension of secret agent to include bankers. But nothing that we know about how the
value of tall depends on the context gives any reason to expect that its value will change
in this way. More generally, nothing we know on independent grounds about the contextsensitivity of such adjectives suggests that they are sensitive to whether a context is
reasonable or unreasonable.22[22]
21[21]
Technically, it is easy enough to formulate such rules. A modification of the rule in
(5) would not be difficult. To highlight our neutrality on just what the semantics of
complex demonstratives must be, we could also give a semantics along the following
lines. [[That]]c=λFλG.Φc(F,G). For contexts like that of (4), where inertness is desired,
Φc(F,G) = G(a(c)), getting the same result as (5) above. For contexts where we do not
want inertness, like that of (6), we could have Φc(F,G) = F(a(c))&G(a(c)). Other options,
more in line with nominal policing, are also available. Note that this semantics is context
dependent, not in the sense that the mode of composition depends on the syntactic
environment, but that it depends on features of the context of utterance.
22[22]
A referee suggested that the fan of the weakened inertness thesis could abandon the
assumption that in reasonable contexts, the rule works with extensions of both modifier
and noun in complex nominals such as tall secret agent. Abandoning this assumption, the
referee suggested, would allow nouns in reasonable contexts to contribute a more general
property (such as the property of being an agent), while adjectives change their
There are other reasons to think this second option makes false predictions. If
sensitivity to whether a context is reasonable or unreasonable is an aspect of the contextdependence of the nominal, then it should appear independently of what determiner goes
with the nominal. With this in mind, consider:
12. The badger is hungry.
In the reasonable context of (4), the second option predicts this will come out true. (It
does not merely say that there is a true speaker meaning associated with the utterances; it
says that the semantic value of the sentence in context is evaluated to truth). This appears
wrong, and is predicted to be wrong on most theories of definite descriptions.
Let us turn to a third way of implementing the weakened inertness thesis.
Whereas the previous option called for the nominal to shift its contribution between
different uses of complex demonstratives, this option calls merely for shiftiness between
nominal contributions in complex demonstratives, on the one hand, and nominal
contributions in other constructions, on the other. It is, therefore, equally a view about the
contributions of that and F in that F. The single sort of contribution allegedly made by
the nominal in complex demonstratives is this. The semantics of that converts the value
of F to reasonably taken to be an F, and the nominal then contributes this value in a
policing role.
As with the previous option, the nominal’s role is not inertness, as posited by the
strong inertness thesis, since the nominal always makes some sort of semantic
contribution. But it satisfies the weakened inertness thesis nonetheless. In reasonable
contexts, That F is G will be true even if the appropriated object is not an F. This will be
so because, by hypothesis of what reasonable contexts are, it is reasonably taken to be F.
The third option may appear to circumvent the problems with the previous proposal, as it
avoids context dependent rules of composition, and does not ask the nominal to shift its
value between uses of complex demonstratives.
Our objection to the third option considers the following pair of sentences:
13.
A. That mouse is hungry.
B. That thing reasonably taken to be a mouse is hungry.
contribution to something innocuous mimicking inertness. The motivation behind this
proposal is to preserve the intuition that the property contributed by whole nominal in
that tall secret agent is simply the property of being an agent.
The point we raised against the second version applies to this one as well. Both
versions have it that the adjective is sensitive to whether the context is reasonable or not,
and nothing we know about the context-sensitivity of adjectives suggests that they are
sensitive to this contrast. For instance, tall is sensitive to a number of contextual features,
both of extra-linguistic context, and of its linguistic environment. But none of this
amounts to sensitivity to whether a context is reasonable or unreasonable. (See Kennedy
1997 for recent work on the context dependence of adjectives like tall.) Even more
generally, where we understand context-dependence, we do not see sensitivity to whether
the context is reasonable or unreasonable.
According to the third option, the complex demonstrative construction makes the
extension of mouse expand to include things (in that context) reasonably taken to be
mice. So it predicts that (13A) and (13B) will be synonymous.23[23]
This prediction appears to be false. Suppose we are in a zoology lecture. The
lecturer informs us that mice and shrews can be reasonably mistaken for one another, and
that even experts sometimes make mistakes. But the lecturer goes on to show us one
animal of each kind, and shows us precisely how to tell by looking which one is which.
She can certainly go on to say (13B) twice in succession, once pointing to the mouse, and
once pointing to the shrew she has just explained can be reasonably taken for a mouse.
Assuming the animals really are hungry, both utterances of (13B) are true. In contrast,
we get no such judgment for the corresponding uses of (13A). There, when she is
pointing at the shrew (which was just explained not to be a mouse), there is a strong
judgment that her utterance of (13A) cannot be true.24[24] More importantly, there is a
strong differential judgment between (13A) and (13B). This is incompatible with the
prediction that they are synonymous.
Let us summarize the dialectic so far. The strong and the weakened inertness
theses are motivated by the imperviousness of communication to (some) misdescription.
The strong inertness thesis has trouble with demonstrative utterances (of the form That F
is G) in which the object appropriated by the use of That F is G, yet in which speakers are
strongly disinclined to assess the utterance as true. The weakened inertness thesis tries to
accommodate these cases, but, we have argued, it cannot plausibly be implemented.
If both the strong and the weakened inertness theses fail, then the remaining
option is that the nominal plays a truth-conditional role of some sort. This would be either
a policing role, or a role whereby an utterance of That F is G is false if the appropriated
object is not F.25[25] We now argue directly for nominal policing.
3. Presuppositional phenomena as evidence for nominal policing
Our argument for nominal policing has the following three-step structure. First, we will
point out a glaring sort of semantic defect to which uses of bare demonstratives (this, that
and their plurals) are prone. In the case of bare demonstratives, it is relatively easy for
hearers to tell when this phenomenon is present, as the defect is something that can be
23[23]
We are supposing that the rule that takes F to reasonably taken to be an F will take
reasonably taken to be an F to itself. The idea is to expand the nominal’s extension in
order to take into account the epistemic situation of the speaker using it. If the epistemic
situation is already accounted for by the nominal, then iterating this process adds nothing.
It is important keep in mind that the rule doesn’t instruct us to add linguistic material.
Adding linguistic material could lead to non-trivial iteration, but that’s not what the
proposal in question offers.
24[24]
As with sentence (6) (That secret agent is hungry), one explanation for speakers’
judgments that the sentence is not true is that speakers have confused lack of truth with
pragmatic inappropriateness. We have already criticized this explanation (in the
discussion of explanation (ii) in our original discussion of (6)), and in any case what
matters here is the difference in judgments about (13A) and (13B), not whether those
judgments are correct.
25[25]
The latter view is defended by Richard (1993), and by Lepore and Ludwig (2000).
readily observed. Aside from the striking feeling of defectiveness, however, there are
also tests, involving the ways in which discourse may sensibly continue, by which one
can detect this phenomenon. The second step of the argument is to present these tests.
In the third step, we move from bare demonstratives to complex demonstratives.
We will argue, using the tests presented in the second step, that certain uses of complex
demonstratives behave the same way as defective uses of bare demonstratives do:
namely, uses in which the speaker appropriates an object that does not satisfy the nominal
turn out (by the lights of the tests) to be just as defective as the defective uses of bare
demonstratives. From the fact that the tests give the same results in both of these cases,
we conclude that the same phenomenon of semantic defectiveness is present both times.
This will amount to an argument in favor of nominal policing. As will become
clear as the discussion progresses, it will offer diagnostics for when an utterance fails to
semantically express a proposition. When an appropriated object fails to satisfy a
nominal, those diagnostics are met, and no proposition is expressed.
3.1 Bare demonstratives
Let us begin with the glaringly defective uses of bare demonstratives. Suppose a speaker
points generally off into the distance and says:
14. #That is a fine piano.
This utterance is defective. There are two important features of its defect. First, it
seriously inhibits communication. Second, the reason it inhibits communication appears
to be somehow semantic—it is not that the utterance is rendered irrelevant, off-topic, or
opaque. In this case, it appears the utterance cannot completely determine the conditions
in which the item referred to by the occurrence of the demonstrative that has the property
of being a fine piano. There is no such object, so we fail to determine truth conditions in
this sense.
In light of this, we shall describe this sort of situation as one in which no
proposition is expressed by the utterance (as we likewise do in nominal-policing). This
formulation is no doubt somewhat theory-laden. The situation could be described in
slightly different (and equally theory-laden) ways. It could be described by saying a
structured proposition is expressed, but is gappy, in virtue of missing a constituent
corresponding to the noun phrase.26[26] Alternatively, it could be described by saying a
false proposition is expressed, but one that is false in a way distinctive of reference
failure. (This is especially theory-laden, as the negation of the sentence would require the
same distinctive falsehood status). No doubt, there are other options.27[27]
In light of the failure to determine truth conditions in the sense we described, we
believe our preferred description is apt. But we also want to stress that for our concerns
here, the theoretical differences between these various descriptions of the situation are
comparatively minor. First of all, the judgment that there is something wrong in cases
like (14) is very clear, and we take it to demonstrate the phenomenon in question,
independently of any description we might choose. Moreover, each description agrees
that the example exhibits some kind of grave semantic defect, be it lack of a proposition,
As noted earlier, this is Braun’s (1994) view. Gappy propositions are further
discussed in Braun (1993).
27[27]
Michael Dummett (1959) discusses this option.
26[26]
incompleteness of a proposition, or a distinctively marked sort of falsehood
corresponding to reference failure.
It will be useful to have a label for the defect at issue. We will call it p-infelicity
(‘p’ for ‘propositionless’). The term ‘infelicity’ is often used to indicate that something is
inappropriate or unacceptable. The judgment that (14) is somehow defective is a clear
case of a judgment of infelicity. There are many kinds of unacceptability, and ‘pinfelicity’ is a label for the sort displayed in the utterance of (14).28[28]
We now turn to the second step of our argument, in which we present tests that
can detect cases of p-infelicity, such as the defect apparent in the utterance of (14) when
the bare demonstrative is empty.
3.2 Tests for p-infelicity
The defective utterance of (14) is an example of a presupposition failure. The utterance
of That is a fine piano presupposes that the demonstrative that indicates a unique
contextually salient individual. It is the failure of this presupposition that leads to failure
to express a proposition. The tests that detect p-infelicity will be tests for a certain sort of
presupposition failure.
Before elaborating the tests, it will be useful to comment on the notion of
presupposition at issue. In linking presupposition to judgments of felicity, we are
working with a pragmatic notion of presupposition. This sort of presupposition describes
the requirements a sentence places upon a context for an utterance of the sentence to be
felicitous in a context. Thus, sentence S presupposes proposition p if (and only if) a
context must satisfy p for an utterance of S to be felicitous in it. A bare demonstrative
triggers the presupposition that the context contain a contextually salient demonstrated
individual, as an utterance of a sentence containing a bare demonstrative will not be
felicitous unless this requirement is met.29[29]
The pragmatic notion of presupposition is to be distinguished from the notion of
logical presupposition: a relation between (gappy) propositions which describes when a
proposition has a truth value. Our notion is pragmatic, in that it is an aspect of the way
28[28]
The notion of infelicity appears in J. L. Austin (1975) as a very general concept of
inappropriateness which might attach to an utterance. Much of the more recent
pragmatics literature, especially the literature on presupposition (e.g. Robert Stalnaker
1974; Lauri Karttunen 1974) concentrates on a more narrow notion of inappropriateness
to assert or utter in a context. Irene Heim (1988) talks about whether a context “admits”
a sentence. Infelicity in this sense remains strictly weaker than, for instance,
uninterpretability. Our notion of infelicity is this latter, more narrow one. However,
there is some reason to doubt that all infelicity even in this more narrow sense is pinfelicity. For discussion of this, see Glanzberg (forthcoming b).
29[29]
This is basically the notion of presupposition of Robert Stalnaker (1974). There is
one important difference, however. We are not assuming Stalnaker’s account of context
and the way contexts satisfy presuppositions. (See note 43 for more discussion.) Also,
note that Stalnaker labels what we have called ‘presuppositions’ as ‘presuppositional
requirements’, as he reserves the term ‘presupposition’ for a propositional attitude used to
describe speaker presuppositions. Our usage is fairly standard among those interested in
how certain linguistic constructions lead to presuppositions.
sentences behave in contexts, rather than a relation between propositions characterized in
many-valued logic. At the same time, it must be stressed that p-infelicity picks out a
subclass of pragmatic presuppositions. Not all infelicities are p-infelicities, and not all
presuppositional requirements lead to p-infelicities when they are not satisfied. We have
argued that bare demonstratives trigger presuppositions which lead to p-infelicity, but we
do not insist that all presuppositions have this effect. Presuppositions leading to pinfelicity are basically expressive presuppositions, of the sort highlighted by P. F.
Strawson (1950): they are requirements placed on a context for a sentence to express a
proposition in that context. Just as p-infelicity is a subspecies of infelicity, this is a
subspecies of (pragmatic) presupposition.30[30]
A presupposed proposition stands in an unusual relation to the sentence that
presupposes it. (When clear enough, we will simply talk about a sentence which
expresses a proposition, and talk about entailments between sentences). Consider a nondefective utterance similar to (14):
15.
A. That is a fine violin.
B. That is not a fine violin.
C. If that is a fine violin, then we should appreciate it.
P: That picks out a unique contextually salient individual.
Each of (A-C) implies (P).31[31] This is a mark of what is usually called the
‘backgroundedness’ of presuppositions: presupposed propositions are background for
felicitous utterances, and so are insensitive to whether it is a sentence or its negation
which is uttered, or whether the sentence appears in a conditional.32[32]
We have noted that bare demonstratives trigger presuppositions that lead to pinfelicity. We will say the same for complex demonstratives. To argue this, we need to
30[30]
The classification in terms of logical, expressive, and pragmatic presupposition is
due to Scott Soames (1989). Glanzberg (forthcoming b) presents examples of
presuppositions whose failures trigger weaker infelicities than p-infelicity.
31[31]
More exactly, if an utterance of A in a context c is true, then it is also true that the
use of that in the utterance picks out a contextually salient individual in c. Mutatis
mutandis for utterances of B and C. In this case, the presupposed proposition is about the
world of utterance. Not all presuppositions have this feature, however. For instance,
consider:
(i)
(i)
A. John regrets voting for Bush.
B. John does not regret voting for Bush.
P. John voted for Bush.
Here (P) is not about the world of utterance, and each of (A-B) implies (P). (In fact, (A)
simply entails (P).)
32[32]
The family of environments in (15) can provide a useful set of diagnostics for
presupposed content. In much of the literature concerned with what contents are
presupposed by what sentential constructions, these diagnostics are often offered as
virtually definitive of presupposition. They also indicate the basic facts about the
projection of presuppositions from their triggers to complex sentences. The literature on
these issues is quite large. For an introduction along the lines we pursue here, see
Chierchia and McConnell-Ginet (2000). More extensive surveys of recent literature can
be found in David Bever (1997) and Nirit Kadmon (2001).
know more clearly how to detect p-infelicity. Many early discussions of presupposition,
notably that of Strawson (1950), linked the relevant notion of felicity to that of truth
value. Strawson noted that in many cases of presupposition failure, especially in cases
like (14), we are not inclined to say that what was said is true, or that it is false. We are
not inclined to assign a truth value. As we mentioned, it is natural to associate this with
the lack of a proposition to be true or false.
Unfortunately, the situation is not quite as clear as this explanation suggests. One
reason is there is often an option to employ a so-called presupposition-canceling
negation. Consider:
16. That is NOT a fine piano—there is nothing there at all.
This is a marked construction, but speakers often have it available. We are left wondering
if the marked nature of the construction allows it to count as assessing a proposition for
truth or falsehood. But nonetheless, in uttering (16), speakers appear to say something
hard to distinguish from declaring an utterance of (14) to be false.33[33]
Luckily, there is some more stable behavior that goes with p-infelicity. Here,
finally, we come to two tests that we will shortly employ with uses of complex
demonstratives. The first involves what we will call echo-assessments; the second
involves indirect speech reports.
An echo-assessment of a sentence S that has already been spoken in a context is a
repetition of S (perhaps correcting for occurrences of indexicals like I and you), preceded
by either yes or no. In responding to an utterance of (14), speakers (who do not have any
misapprehension that there is a demonstrated object) are strongly unwilling to make
echo-assessments. They will not say either of:
17.
A. #Yes, that is a fine piano.
B. #No, that is not a fine piano.
When they do offer a negative judgment, as in (16), they do so by way of initiating a
repair, marked by the stressed negation and the gloss on what semantic defect is present
in the original utterance. This is the first test for p-infelicity. If an utterance is pinfelicitous, then echo-assessments are unacceptable, without initiating a repair.
The second test involves indirect speech reports. Speakers typically will not offer
indirect speech reports that make use of the defective phrase. In response to (14), for
example, they will not say simply:
18. #He said that is a fine piano.
If they do attempt this, they will typically initiate a repair, as in:
19. He said ‘that is a fine piano’, but I don’t know what he was pointing at.
That is the second test for p-infelicity. If an utterance is p-infelicitous, then indirect
speech report of it is unacceptable, without initiating a repair.
A crucial point about the repairs in these cases is that they are obligatory. The
conversation cannot be acceptably continued, in these cases, unless some repair is made.
Before speakers in the discourse can comment on the truth of p-infelicitous utterance, or
report its content, the utterance must be repaired. Unlike the truth-value test, the need for
33[33]
It may well be that these turn out to be what Laurence Horn (1989) calls
‘metalinguistic negation’. We are not committed to any particular analysis of this
phenomenon here; we only need to note that constructions like (16) make it difficult to
apply the Strawsonsian test for p-infelicity.
repair is not made difficult to detect by such devices as presupposition-canceling
negation. That just is a repair strategy. We propose that these two tests offer a reliable
guide to p-infelicity.
Each of the tests we have proposed targets a central aspect of the notion of
proposition expression: propositions are bearers of truth, and propositions are what
speakers say. The echo-assessment test targets a proposition expressed by an utterance as
the bearer of truth, which determines if the utterance is correct or not. If an echoassessment fails, in a given context, it is because an assessment for truth of what was
explicitly said by a speaker cannot be given, without initiating a repair of their attempt to
express something. Hence, their attempt was sufficiently defective to preclude truth
assessment. This, we suggest, is a central aspect of their having failed to express a
proposition at all. This test is thus a more refined version of the traditional test of truth
value judgments.
The indirect speech report test targets the second aspect of proposition expression:
propositions are what speakers say. If a speaker cannot give an indirect speech report, it
is because she cannot appropriately express something the initial speaker might have
been attempting to say by the means that speaker used. This occurs if in fact the first
speaker’s attempt to express a proposition—to determine ‘what was said’—failed. As
both features—being bearers of truth and being what speakers say—are aspects of the
single notion of proposition expression, the tests are typically passed or failed together.
It is important to stress that information from a p-infelicitous utterance may
nonetheless reach a knowledgeable interpreter. There are many cases in which
information is conveyed which are clearly not cases of expressing a proposition. Consider
a sun-burnt man, parched and dirty, who crawls out of the desert, pointing to a pitcher of
water and attempting to speak. Too parched, he merely manages to produce a scratchy
sound vaguely like:
20. Waa…
By nearly anyone’s lights, no proposition is semantically expressed.34[34] But lots of
information is available in the act. The act conveys that he man is thirsty, wants water,
wants that water, wants it more than to call his wife, etc. Information may be conveyed
in all kinds of ways, many of which do not rise to the level of expressing propositions.
There is not the space here to explore fully what this difference amounts to.35[35]
But let us simply note that it is important that the issue is one of expressing a proposition,
by the assertion of a specific sentence, in a specific context. That can fail, even if there is
a proposition that might be charitably attributed to the speaker. This comes out, for
instance, when we apply the indirect discourse tests. There may well be cases where the
reporter knows fully well what the speaker meant to convey, and can in many cases
report it. To apply the test, we need to consider whether the reporter can report what the
speaker meant, using her very words (allowing certain small modifications, like changing
I to she). It is this which targets whether a proposition was expressed, rather than merely
whether information was made available.
34[34]
An exception may be Robert Stainton (1995), who has argued that there are cases of
non-sentential assertion; but even Stainton may well not hold that a proposition is
expressed in a case like (20), which fails to have even a single lexical item.
35[35]
Glanzberg (forthcoming a) explores it at greater length.
We have said that the relevant notion of repair is obligatory repair. This is a
normative notion. The norm in question is a norm of discourse, as is appropriate for one
that tests for presupposition. Such norms do not operate in a vacuum. If we concoct a
case in which a speaker can save a small child from a horrible fate by making an
otherwise bad indirect speech report, we may rest assured most decent speakers would do
it without a second thought. A norm of discourse brings with it a felt unwillingness on
the parts of speakers, and a tendency to seek out ways to avoid violations. Repair
initiation is a well-established way to do so.
The basic mark of p-infelicity, and the presuppositions which induce it, is
obligatory repair. Our tests for p-infelicity are empirical tests, and to give reliable results,
they must be applied in good experimental circumstances. There are two dimensions to
good experimental circumstances. They concern the epistemic situation of the speakers
on the one hand, and their overall normative situation on the other.
First, in good experimental circumstances, the speakers are epistemically wellsituated, so that there is no question about what sentence is uttered, or what the values of
its constituents are. If you're behind a curtain and I say Look, he’s swallowing fire!, this
is not a good circumstance for you to test for whether I have expressed a proposition.
What is needed is that you be epistemically well-situated to assess the value if any of the
constituents, and in this case, you are not able to assess the value of he. Moreover, it
must have been a fully grammatical sentence that was uttered, since a grossly
ungrammatical sentence will also require repair, but not for reasons of p-infelicity.
(Likewise the utterance must not display any gross phonological or phonetic defects, etc.)
Another factor in being epistemically well-situated is that speakers know the rules
for when to change words in belief reports, indirect speech reports, or echo-assessments.
The rules are clear for when to change I to she or you, for instance. Hence, if B reports
A’s utterance of I’m hungry, B will say, Yes, you’re hungry. 36[36]
Second, experimental circumstances isolate the relevant discursive norms. In
general, to see whether a norm is operative, the influence of competing norms must be
screened out, lest they mask the fact that the norm being tested for is violated. In the case
of our tests, a very good experimental circumstance is the courtroom setting, where a
witness is being asked by a lawyer for an echo-assessment or an indirect speech report. In
the courtroom (as we imagine it), there is a clear premium on speaking correctly in the
most narrow sense. For the witness on the stand, whether or not what they say is
misleading, unhelpful, or irrelevant, is not really their concern. Their obligation is only to
speak correctly.
In fact, the courtroom setting is a good experimental circumstance along both
dimensions. It makes vivid the assumption that all sentences are well-formed, and all the
information needed to determine the values of their constituents is common knowledge.
It also makes vivid the screening-off of competing norms. When the tests bear in these
cases, they all the better detect the basic matters of whether truth-evaluable information
36[36]
Note that it is not definitive to being in good experimental circumstances that one
can echo-assess and make indirect speech reports without having to repair. If this were
so, then the tests would be useless, since what we wanted them for to help us diagnose
when repair is obligatory, and so when there is p-infelicity. There seems to be an intuitive
notion of epistemic well-situatedness, as illustrated by the curtain case.
was expressed by the very utterance in question. They all the better detect whether a
proposition was expressed.
To summarize, when they are applied in good experimental circumstances, our
tests provide necessary and sufficient conditions for p-infelicity. But the following
caveats must be kept in view. First, the sufficient condition is obligatory repair in the
absence of other grounds for repair, such as ignorance of context, phonetic defects, etc.
(We might add syntactic or phonological defects as well, but these may amount to failure
to produce a well-formed sentence at all, which certainly counts as failure to express a
proposition.) Second, the tests allow us to gather empirical evidence for p-infelicity,
when they show that speakers take repair to be obligatory in certain settings. As the force
of obligatory repair is normative, speakers may not act upon it, if other norms are
considered stronger in some occasion. Hence, we proposed, the repair tests should be run
in an appropriate experimental setting, such as the courtroom setting. In short, the tests
are well tailored to gather evidence for p-infelicity. But they remain tools for gathering
evidence, rather than providing an analysis of the phenomenon of failing to express a
proposition. When properly applied, the tests detect p-infelicity, and are in this way a
sufficient condition.37[37]
We now turn to the third step of our argument: applying the tests for p-infelicity
to uses of complex demonstratives.
3.3 P-infelicity in complex demonstratives
Earlier, we discussed the notion of appropriation. Demonstratives, both bare and
complex, require a supplement from the context. In the case of classic perceptual uses of
complex demonstratives, the need for a distinctive contextual supplement is illustrated by
the existence of contexts in which an utterance of (1) That key is bigger than that key
would be felicitous, but (2) The key is bigger than the key would not. We have been
neutral on what the nature of this supplement is, offering no more than this functional
characterization. But as we discussed in Section 1, some candidates include speaker
intentions of some sort, and publically accessible gestures of pointing.
One route to p-infelicity in complex demonstratives is through failure of
appropriation. Call this the pre-appropriation route to p-infelicity. This sort of failure
leads to genuine p-infelicity. Suppose, for instance, I point at an empty counter and say:
21. #That key is mine.
37[37]
The comparison between our tests and empirical tests in other areas is
straightforward. Suppose we have a test for the presence of some virus in the blood.
Suppose the test is difficult to apply. Say it will yield false positives if the sample is
contaminated in any way, and false negatives if the sample is drawn from the wrong part
of the body. We would certainly conclude we would like to find a better test—one that is
easier to apply in the field. But we can still say we know a necessary and sufficient
condition for concluding someone has the disease. Likewise, our tools for gathering
evidence for p-infelicity are not perfect, as they can produce false negatives or false
positives if applied in contaminated conditions. And for a normative notion like
obligatory repair, a fully sterile experimental condition may be impossible to find.
Nevertheless, evidence that speakers will judge repair to be obligatory is strong evidence
for p-infelicity.
I have failed to appropriate anything by my use of that key. Compare this with the case in
which I say:
22. #That is mine.
Both utterances are similarly defective. A speaker will not provide an indirect speech or
an echo-assessment for either (21) or (22) without initiating a repair. For example,
applying the indirect speech report test to (21) gives:
23.
A. #She said that key is hers.
B. She said ‘that key is mine’, but I don’t know what she meant, because
there was nothing there.
The pre-appropriation route to p-infelicity really does lead to p-infelicity. Even so, as
pre-appropriation failures are not generally due to the nominal, these are not the crucial
cases for policing.
It is the second route to p-infelicity that is especially relevant to our defense of
nominal policing, so we will focus on it. This route is through appropriation success,
where there is an object that the speaker appropriates by her (classic perceptual) use of a
complex demonstrative, but it fails to satisfy the nominal. This would be a postappropriation route to p-infelicity. To argue for policing, we have to argue that this
really is a route to p-infelicity as well.
Suppose someone at a restaurant is speaking to the waiter. The silverware is in
plain view to everyone. Pointing clearly to the fork, the speaker says:
24. #That knife is dirty.
This is indeed a p-infelicity. Running our diagnostics, we can see that the waiter will be
unwilling to offer an echo-assessment. He will not say either:
25.
A. #Yes, that knife is dirty.
B. B. #No, that knife is not dirty.
Nor will we find indirect speech reports like:
26. #The person at table 8 said that knife is dirty. [Speaking to the cook.]
The speaker might avoid the issue, by saying:
27. Yes, that [pointing] is dirty.
Or perhaps a repair might be initiated, with:
28. Yes, I see your FORK is dirty. I’ll get a new one.
As we have described the case, there is an appropriated object: the fork. We relied upon
the pointing gesture to make this clear, but it is not crucial to the case. We might well
have appealed to intentions manifested in some other way. But however appropriation
happens, there is further reason to think it happens in this case. The further reason is
illustrated by non-linguistic behavior. Suppose we have a command following (24), as
in:
29. #That knife is dirty. Bring me a new one!
We have marked infelicity here. It is p-infelicity, as our tests will readily show. But a
thoughtful waiter, trying to satisfy his client’s desires, will still have no trouble picking
up the fork and replacing it. This shows we have appropriation, though we still have pinfelicity.
What we see here is evidence for nominal policing. The failure of an appropriated
object to satisfy the nominal leads to p-infelicity. Failure to satisfy the nominal is thus
tantamount to failure to express a proposition. So there is proposition failure in cases in
which an appropriated object fails to satisfy the nominal, which is just what nominal
policing predicts. Moreover, the requirement that the appropriated object satisfy the
nominal behaves precisely as we expect of a presupposition which triggers p-infelicity.
We see just as much:
30.
A. #That knife is dirty.
B. #That knife is not dirty.
C. #If that knife is dirty, I will make a scene.
We thus conclude that it is a presupposition of sentences with complex demonstratives
that the appropriated object satisfy the nominal, and if this presupposition fails, we have
p-infelicity. This is nominal policing.38[38]
It may be objected that if there is appropriation, then the sentence uttered in fact
determines a proposition in context. On this view, and utterance of That F is G is true
just in case the appropriated object is G. This objection is in effect to deny the data we
have adduced in (25-29).
As we discussed in Section 3.2, the way to bolster our data is to pay closer
attention to experimental circumstances, by looking at the ‘courtroom’ setting. Consider
the ‘courtroom’ version of (24). The lawyer holds up a fork and says to the witness on
the stand:
31. Yes or no: this knife killed the victim.
In this case, the witness will find repair obligatory in echo-assessment. We see:
32. #Yes, that knife killed the victim.
The objection is presumably based on the observation that we can have:
33. Yes, that killed the victim.
This is perfectly acceptable; but it is not an echo-assessment.
There is a strong differential judgment between (32) and (33). (33) is entirely
acceptable, while (32) is markedly bad. If the objection where correct, then given that we
have clear appropriation of the same object in both, we would expect little or no
difference between the two. With policing, we predict just the difference we see.
It is worth stressing here that the felicitous (33) constitutes an understated repair,
which substitutes a successful bare demonstrative for a defective complex demonstrative
in answering (31). A more pronounced repair would be:
34. Well, that is a fork, but that fork did kill the victim.
38[38]
Because of the behavior of presuppositions under negation, we do not follow
Richard (1993) and Lepore and Ludwig (2000), which predict that That F is G has the
same truth-conditions as That is F and G (where the demonstrative in both cases is a
device of direct reference). This prediction requires that an utterance of That F is not G
be true when the appropriated object fails to satisfy F, while we maintain it is a pinfelicity. A slightly more delicate matter is the proposal of Dever (2001), that complex
demonstratives map to non-restrictive relative clauses, so That F is G has the truth
conditions of That, which is F, is G. Judgments about these cases are not clear, but with
Dever himself, we do not see false non-restrictive relative clauses as inducing pinfelicity. We take this as a reason to doubt that complex demonstratives map to such
clauses. We hence believe our proposal is more accurate, and simpler, than Dever’s. (Cf.
Chierchia and McConnell-Ginet (2000) for an argument that non-restrictive relative
clauses do not induce presuppositions.)
We thus find, as policing predicts, p-infelicity with the complex demonstrative, but not
the bare demonstrative.
It might also be objected that defective utterances containing complex
demonstratives such as (24) are p-infelicitous for a different reason than the defective
uses of bare demonstratives are. If this is right, then p-infelicity is not the unified
phenomenon of failing to express a proposition that we have presented it to be. For
instance, it might be suggested that what is bad about (32) is more accurately glossed as
its being misleading. But this is at odds with the obligatory nature of the repair. Suppose
the witness is a doctor. The doctor knows the fork pierced victim’s heart, but by some
medical peculiarity, its piercing did not kill the victim. Rather, what killed him was a
long-standing liver disease resulting from years of desultory living. The prosecutor may
ask (35A), pointedly, and be answered by (35B):
35.
A. Yes or no: this fork pierced the victim’s heart.
B. Yes, this fork pierced the victim’s heart.
Here, the echo-assessment (35B) is highly misleading, as the doctor fully knows. But the
repair is not obligatory, and in this setting, (35B) is in no way defective. There is a repair
that the doctor might make, to cancel the misleading effect, but it is optional. Optionally,
the doctor can say:
36. Yes, this fork pierced the victim’s heart…but that did not kill him.
The courtroom situation highlights the optional nature of this response. It is unlikely the
doctor will make the repair (or be allowed to make it) in this setting.
A closely related objection is that the acceptability of (33) shows that the content
of lawyer’s question all along was what follows yes in (33). We disagree. What the
acceptability of (33) shows is simply that, as in more extreme cases like (20), there is a
way to express what appears to be the intent of the lawyer’s question. The report in (33)
is quite easy to come by, and quite close to the words used by the lawyer, but it remains a
different sentence, even if close. It remains a repair. That a different sentence can be an
accurate representation of what the lawyer had in mind is no more significant than
extreme cases like (20).39[39]
39[39]
There are a number of difficult cases of applying the tests which deserve comment.
We will restrict ourselves to one, suggested by Jeff King.
Suppose we are in a knife factory, so salient knives are everywhere. Suppose also
there are a few salient forks as well, and a speaker says:
(i) All the forks are dirty.
Finally, suppose there is enough public indication of his intentions that it is clear the
speaker intents to talk about the knives.
Intuitively, we would expect no p-infelicity in (i). The domain of quantification is
not empty, and the quantifier in (i), it seems reasonable to assume, does not trigger such a
presupposition. Nonetheless at least one of our tests is inadvertently tripped. Speakers are
at least somewhat unwilling to give echo-assessments in cases like this: some of our
informants found it unacceptable to say either:
(ii) A. ?Yes, all the forks are dirty.
B. ?No, all the forks area not dirty.
On the other hand, many of these same people find it fine to give indirect speech reports,
with an optional repair, as in:
There may remain a worry that the problem in (32), and the other policing
examples we have considered, do not ‘feel’ like those in bare demonstrative cases like the
original (14). They do not ‘feel’ like there is no proposition expressed. We suggest that
the appearance of a proposition is the misleading result of the ease of the switch from
(32) to (33), which obscures the difference, highlighted in (20), between expressing a
proposition and making information available. This may return us to the issue we have
already discussed, of whether or not to insist that p-infelicity is to be glossed as ‘failure to
express a proposition’ or some other grave defect. But as we said, the two p-infelicity
tests target the basic features of the phenomenon in question, whatever its name. In pinfelicitous uses of complex demonstratives, truth value cannot be straightforwardly
(iii) He said all the forks are dirty ... but that was not really what he meant.
In light of this case, one might worry that what holds here might hold as well in
the case of complex demonstratives—and that would undermine our case for finding pinfelicity when the appropriated object does not satisfy the nominal.
We think that what is important about this sort of case (if these judgments indeed
hold firm) is that the two tests come apart. Above, we gave reasons to think the two tests
would be passed or failed together, yet here we see a case where one is passed and the
other failed. We suspect that what the tests coming apart indicates is that there is too
much 'noise' interfering with the tests, rendering what might have seemed to be adequate
experimental conditions not good enough. (As with our imagined blood test in footnote
37, we have always granted that applying the tests properly can be tricky.) The 'noise' in
this case would be the manifest uncharitability of echo-assessing what the speaker clearly
did not mean. This might in effect violate a competing norm, and hence induce a conflict
between norms that renders the experimental circumstances unfavorable. (The situation
is probably more complicated, though. As Gregory Ward pointed out, there are a number
of ways that we can find noise in echo-assessment, depending on just what the speaker
takes the point of assessment to be, what the standard of correctness is, etc.) Some
speakers do not find this noise interrupts indirect speech report; perhaps because they are
simply passing the statement on, and not called upon to evaluate it themselves.
Interestingly, we see something similar in another sort of case we do not think yields pinfelicity (though we do not hazard a proposal about what they do): non-restrictive
relatives. Consider:
(iv) Arnold Schwarzenegger, who is a weakling, is governor of California.
Indirect speech reports of (iv) are easy; yet many find it difficult to echo-assess. Here the
problem is not so much a norm of charity, but simply not understanding how the frame
of echo-assessment applies to such sentences. (A similar sort of case involving
implicatures from sentential connectives was suggested by Mandy Simons.)
So long as we can observe such noise by the tests coming apart, we maintain this
is not a worry. Though it restrict the cases in which our tests easily yield positive
answers, it does not undercut the claim that when they do give such answers. We have
argued that there is enough such evidence that the tests support policing, when coupled
with the arguments against inertness.
calculated, and what the speaker said cannot be straightforwardly reported. This is
evidence for nominal policing.40[40]
4. Communicative smoothness and policing
So far, we have argued that the strong and the weakened inertness theses are
unsustainable, and we have presented evidence in favor of nominal policing. Together,
these make a strong case: we must have some policing, and the only way to have it
sometimes is to have it all the time.
But there remain a few cases where policing appears to be too much: the very
cases that motivated the inertness theses to begin with. Consider the earlier example (4),
repeated here:
4. That badger is hungry.
40[40]
As empirical means for testing for presupposition, the tests for p-infelicity can be
applied to a variety of utterances. A referee raised the question of what results they would
give when applied to utterances containing definite descriptions in which the existence
condition or the uniqueness condition (or the combination) is not met. The referee
objected to our use of the tests on the grounds that (a) the tests deliver the result that
utterances of the form The F is G, where one or both conditions are not met, are pinfelicitous; but (b) existence and uniqueness conditions of definite descriptions should
be treated in the way that Russell proposed, and thus are not presuppositional. Armed
with assumption (b), the referee concluded that the correct verdict on such utterances is
that they are false, and thus objected that the tests do not test for p-infelicity after all.
We regard it as an open empirical question whether (a) is correct. If it is, then it is
strong evidence that definite descriptions carry presuppositions. This contradicts the
classical Russellian view (b), though it by no means precludes an analysis where
descriptions are restricted quantifiers. (Likewise, it does not imply the classical
Strawsonsian view according to which they are simply devices of reference, on par with
other such devices.) The presuppositional status of definite descriptions is still widely
debated. Firm evidence on (a) would help to resolve this issue, but we are uncertain
whether there is such evidence. We thus remain neutral on the status of definite
descriptions.
We do not, however, remain neutral about the tests. The tests are a means to
gather data, and we offered reasons that they provide data about p-infelicity (viz. the tests
target the target the two central roles of propositions: to be bearers of truth-value, and to
be what speakers say). We are committed to following the data where it leads, even if
this contradicts a Russellian view of descriptions. What would be required not to do so
would be a counter-argument that, in spite of our reasons, the tests do not really test for
p-infelicity after all.
Though we remain neutral about the ultimate status of definite descriptions, our
tests do tell against the position which holds that definite descriptions do not carry
presuppositions, and hence do not lead to p-infelicity; and that the correct semantics for
complex demonstratives assimilates them to definite descriptions. (Such a position is
defended by Lepore and Ludwig (2000). This combination of views is incompatible with
what we have argued here, and our defense of the tests is an argument against it. (See
note 38 for additional arguments against Lepore and Ludwig’s position.)
In the context we described, the fact that the appropriated animal is not a badger, but a
fox, does not appear to cause any evident problem. This might tempt one to embrace an
inertness thesis. Though we have argued against inertness theses, we would now like to
explain why the cases motivating them are not counterexamples to nominal policing, and
why they don’t support inertness theses as much as they might seem to initially.
Let us begin with a new example. Suppose I am highly knowledgeable about
cooking, sharing a kitchen with someone less knowledgeable who confuses colanders and
sieves. Pointing to a sieve, she says:
37. The holes in that colander are too small.
I know that the appropriated object is not a colander. So as far as I am concerned, the
utterance is markedly odd. But there is an extraordinary measure I can take. I can
recognize that my companion has little if any idea what she is talking about, and is
misusing the word. I can play along, and even reassure my companion by saying:
38. No, the holes in that colander are just fine.
If I say (38), there will be no evident problem with communication. Indeed, I might
choose to say (38) for just this reason: communication will proceed as well as it needs to
for the purposes at hand if I do.
The communicative smoothness enabled by my utterance of (38) is compatible
with p-infelicity. If I play along in this way, pretending to have the same mistaken belief
as my interlocutor, I will experience the discomfort that comes from knowingly misusing
a word. Although the mistaken interlocutor may not feel it, there is something defective
about both utterances (37) and (38). The defect would show up if someone else, who was
neither ignorant nor pretending to be, reported the conversation. Suppose, for instance,
that I tried to report what my mistaken interlocutor said to someone else. I would not say:
39. #She said that the holes in that colander are too small.
I would not say this, because I would be following the linguistic norm that I suspended
for the sake of communicative efficiency in uttering (38). Even in that context, repair is
obligatory. The speaker who utters (38), while knowing that the appropriated object is not
a colander, in effect trades off the obligation to follow one linguistic norm for the sake of
following a competing norm. In this case, the norm of efficient communication wins
over the norm obligating repair.41[41]
What about the original fox/badger case (4)? There too, speakers can make do
with the utterance of (4), if their epistemic state is such that they cannot distinguish foxes
from badgers. They can then simply assume that the animal is a badger, and go on
thinking that things are fine. And even if they know too much about what badgers look
like to assume this (badgers do not really look like foxes!), speakers may still find it
possible to carry on as if the animal was a badger—just as one might, in response to (37),
speak as if sieves were called ‘colanders’. The crucial point is this: the fact that locally
41[41]
Our guidelines for applying the repair tests tell us to seek good experimental
conditions, such as the ‘courtroom’ setting, where the norm efficient communication is
not so prominent. In such a setting, an utterance of (37) would indeed trigger a repair.
Suppose the interlocutor is a lawyer, who demands of the witness under oath:
(i) Yes or no: Are the holes in that colander too small?
In such a context, unlike the easygoing one described above, the obligatory nature of the
repair is made manifest.
there might be conversational smoothness does not show that there is not a problem with
the discourse that would appear in other settings. The norm which obligates repair may
be in effect, even if we do not follow it because we are placing a premium on
communicative smoothness locally. Local conversational smoothness is not enough to
avoid p-infelicity.
Another way to see this point is to consider an extreme case. Suppose two people
simultaneously hallucinate a green monster, which seems to both of them to be in the
same location. One of them points in a way that would appropriate the monster if it were
really there—and in a way that appears to both hallucinators to do just that—and says:
40. #I am frightened of that.
For the hallucinators, communication is entirely smooth. But it is a rather drastic step to
say that a proposition is expressed here. The context is defective enough to refuse to
count a proposition as expressed even if the speakers think there is one, and even if it
appears locally that they communicate smoothly. That is a reason to think that there is pinfelicity, despite how things seem to the speaker and hearer.
The moral to be drawn here is that not all participants in a conversation will be
able to recognize a p-infelicity, and that repair does not cease to be obligatory, simply
because there is communicative smoothness. In some cases, the presence of p-infelicity is
not enough to impede all aspects of communication. One can get what one needs out of
the conversation, even if there is a problem that would become evident in a wider
context.
We have argued that the cases motivating inertness theses are not
counterexamples to nominal policing. We have not said very much, however, to explain
the intuition that the utterances in these cases are true, other than pointing out that they
can be met with communicative smoothness, rather than attempts at repair. Much more
can be said about what underlies communicative smoothness in these cases, and this will
be our focus for the rest of the section.
In (38), the object appropriated by a use of a complex demonstrative does not
satisfy the nominal, and the speaker knows it. For cases like these, there is a natural story
to tell about the communicative smoothness in terms of the dynamics of conversations.
Let us begin with some preliminary remarks.
We think of conversation as a matter of exchanging information. Following
Stalnaker (1974, 1978) and David Lewis (1979), we think of these exchanges as building
a conversational record of information that has been accepted in the course of the
conversation, together with whatever background information may also be taken for
granted by participants in the conversation. Information is normally added to the record
by assertion. If the content of an assertion is accepted in the conversation, it is added to
the conversational record.42[42]
Whatever is in the conversational record becomes common ground among
speakers in a conversation. It is common also to think of a context in terms of the
42[42]
Of course, this is an idealization in many ways. Not all conversations are oriented
towards building a record of information. Those that are involve more structure, such as
a complex structure of turn-taking. But none of these issues affects our point about
conversational smoothness and policing. (For a survey of some of these sorts of issues,
see Levinson 1983.)
accumulated common ground information at a given point in a conversation. The context
of an utterance may be thought of as the current common ground—the current state of the
conversational record—at the point of utterance. From this perspective, for instance,
what the indexical you refers to is a matter of who is taken to be the addressee by
speakers, which is a matter of what the common-ground information about the addressee
is. We might think of appropriation in these terms as well. What object is appropriated
by an utterance might be a matter of what common-ground information about the salience
of objects is present in a given utterance. (Though as we discussed in Section 1, we do
not adopt this as our official view, for it may preclude a more extensive role for speakers’
intentions. As we said there, we wish to remain neutral on this issue.)
It is well-known that there are other ways to add information to a conversational
record than assertion. One that is especially important is accommodation. Consider the
familiar case of a factive verb:
41.
A. John regrets voting for Reagan.
B. John does not regret voting for Reagan.
P. John voted for Reagan.
The presupposition of (41A) is (P). If the information that John voted for Reagan (P) is
not in a given context—not part of the conversational record—an utterance of (A) may
appear infelicitous. It may. But more often, speakers will simply add this information to
the context, and move on. This may well induce a sort of ‘double-take’ feeling, but it
will not lead to any genuine infelicity. This is the process of accommodation. As we are
describing it, accommodation is a kind of Gricean, on-the-fly repair strategy. To avoid
rendering an utterance infelicitous, speakers add the information required by its
presuppositions to a context, and proceed as if that information had been there all along.
A rather stylized version of this can be seen with clefts, as in the textbook that starts:
42. It was in 1812 that Napoleon fought the battle of Borodino.
Such an utterance in a somewhat less direct way conveys the date of the battle.43[43]
Presupposition accommodation can happen with complex demonstratives as well.
Suppose my car mechanic tells me Your fuel injector is cracked, while pointing to a
silvery thing in the engine of my car. Some time later, I am complaining about this, and
to make the point, I open the car’s hood, point down to the silvery thing and say:
43. That fuel injector is cracked.
This can be perfectly felicitous. Suppose the context provides an appropriated object.
Let us suppose, for instance, that my pointing is clear. Suppose, furthermore, that I did in
fact glean the right thing from my mechanic, and what I point to really is a fuel injector.
43[43]
We are treating this sort of accommodation as type of conversational repair,
inducing what is sometimes glossed as the phenomenon of ‘informative presupposition’.
This is fairly close to Lewis’s original use of the term, and may well be close to
Stalnaker’s treatment as well (see Simons 2003 for more discussion). In more recent
literature, such as van der Sandt (1992), accommodation is more closely allied with
projection mechanisms.
Factive presuppositions are discussed at length by Stalnaker (1974). Clefts have
been the subject of much discussion, including Atlas and Levinson (1981), Delin (1992),
and Glanzberg (forthcoming b).
(It could easily have been otherwise. My mechanic could have been pointing to the place
where the gasoline spilled out, rather than the fuel injector.)
In this case, both the speaker and hearer have accommodated. Each adds to her
store of information that the thing pointed to is a fuel injector. (No doubt, both have
highly partial understanding of this proposition, but they can add it nonetheless.) The
result is that this proposition becomes common ground among speakers. It becomes
information in the context. In the fuel injector case (43), the accommodated proposition is
in fact correct. But this is not necessary for the process. The same phenomenon is at
work in the colander/sieve case (38), when the speaker knows that the appropriated object
is not a colander but for present purposes decides to speak as if it is. The speaker might
have been inclined to take what colanders are as common ground, in a cooking setting.
But she might now revise this supposition, adding information in the common ground
that she knows to be false—namely, that the object in question is a colander. There is no
general requirement that information in a conversational record be true. This is asking too
much. Nor is there even a requirement that all speakers believe it. It is possible to go
along with a conversation, even if one does not believe what is said.
Summing up, the communicative smoothness in the cases motivating the inertness
theses is explained as follows. What appears to impede smooth communication, as far as
speakers in the conversation are concerned, is the information contained in the context.
Impairment can be avoided by such strategies as accommodation, which is a way of
adjusting this information. We have argued that this can take place in spite of grave
semantic defects in an utterance: in spite of p-infelicity. Thus, if the situation is right, the
norm of repair can be overridden to preserve communicative smoothness. But, we have
argued, there is p-infelicity nonetheless.44[44]
44[44]
Throughout, we have treated nominal policing as the thesis that the object
appropriated by a use of a complex demonstrative must satisfy the nominal, in order for
the utterance in which it occurs to express a proposition. We have argued that this thesis
is true in virtue of the presuppositional behavior of complex demonstratives. Some
linguists and philosophers (such as Stalnaker 1974, 1978) hold a view of presupposition
that would in effect recast nominal policing in epistemic terms. This view builds on the
idea of context as common-ground information. According to it, the only issue for
satisfaction of presuppositional requirements is whether the information in a context
entails a presupposition. (Propositional presuppositions, such as those of factives like
(41), are naturally treated this way.) If we treated policing this way, it would be recast as
an epistemic thesis: a nominal would count as satisfied in a case in which all speakers
accept a mistaken proposition.
In explaining problematic examples like (4) and (37), we have in effect noted that
this more epistemic approach is useful in explaining phenomena of conversational
smoothness. But we have also noted that in some cases, this smoothness is the result of a
kind of pretense. In (39), for instance, the accommodating speaker finds uses of that
colander marked, and cannot make them outside of the highly localized context.
Though we have not argued against this more epistemic view as a basis for
approaching the wide range of presuppositional phenomena, we have in essence argued
that it is not adequate for the presuppositions of demonstratives. We think this is fairly
clear in hallucination cases like (40). More generally, we have suggested that even when
5. Conclusion
Complex demonstratives in their classic perceptual uses are standardly taken to be
paradigmatic referring expressions. If the nominal in such uses of complex
demonstratives plays a policing role, this does not mitigate against the standard
categorization of complex demonstratives as referring expressions, as other defenders of
nominal policing have pointed out (e.g. Braun 1994; Borg 2000).
We said at the start that our defense of nominal policing, unlike previous ones,
leaves it open whether complex demonstratives are devices of reference or not. But it
may initially seem mysterious how they could fail to be such devices. How, for example,
could complex demonstratives plausibly be thought to be quantificational?
Jeffrey King (2001a. Ch. 2) has recently defended a view according to which, in a
context of utterance, that in that F contributes a two-place relation between properties to
the semantic value of utterance in which it occurs, just as the contributions of some,
every, one, and other generalized quantifiers are thought to do. (It is not unproblematic
what makes an expression a quantifier; King points to several syntactic and semantic
features that complex demonstratives, on his view, share with paradigmatic quantifiers.)
According to King, the lexical meaning of that in that F—its meaning, considered
outside of any context—is a function from properties and context to a truth value. Where
blanks are placeholders for properties, King thinks that the proposition expressed by an
utterance of That F is G is structured in the following way:
44.
__ and __ are uniquely __ in an object x and x is __.
The first and last blank are filled in by F and G, respectively. How the middle two blanks
are filled depends on whether the use of that F is rigid or not.
We have been concerned exclusively with classic perceptual uses, which are rigid,
so let us focus on those. When the use of that F is a classic perceptual use (and so is
rigid), the second blank is filled with the property of being identical to the appropriated
object, and the third blank is filled with the property of being jointly instantiated in the
context. The second blank is needed, King thinks, to account for what we have called
appropriation. Call this blank the proposition's appropriation position.
Let us see what a simple example looks like on King’s view. Suppose someone
utters:
45. That fox in the garbage is hungry.
Suppose the object appropriated by the use of that fox in the garbage is the creature
Frida. Then the property filling in the appropriation position is the property of being
identical to Frida, and the entire proposition, according to King, will be:
46. FOX IN GARBAGE and =FRIDA are uniquely jointly instantiated in c in an
object x and x is HUNGRY.
the participants in a given conversation do not recognize a defect in their commonground information, and so do not recognize p-infelicity, it may be present nonetheless.
As we noted in discussing (39), the infelicity may still be brought to light in wider
conversational settings, where speakers not entertaining the same propositions are
involved. Generally, though speakers need not believe all the propositions in a
conversational record, gross errors in the record, or asymmetry in speakers attitudes, can
constitute defects in a context which can generate p-infelicity.
(Here c stands for the context of utterance, and capitals indicate properties.)
Now, with respect to nominal policing, King’s account is silent. King’s account
leaves open what happens if there is no unique thing that is both Frida and a fox in the
garbage. It leaves open whether an utterance in such a context is straightforwardly false,
or fails to express a proposition. But the latter option could easily be written into King’s
semantics, without spoiling the status of the expression as a quantifier. This would simply
be a matter of appending a clause that says there is a presupposition that there is a unique
object that has both F and the property in the appropriation position.45[45]
If such a presupposition were added to King’s semantics, the result would be a
quantificational account of complex demonstratives in which the nominal plays a
policing role.46[46] This illustrates the neutrality of our defense of nominal policing. If our
defense works, then complex demonstratives (at least when used in the classic perceptual
way) carry a presupposition that the appropriated object satisfies the nominal. But the fact
that these uses carry such a presupposition places no constraints on whether the
expression is a device of direct reference, a device of reference of some other sort, a
quantifier à la King, or something else.47[47] The presuppositions that we have argued
complex demonstratives carry do not map on to any particular semantic category.
Suppose that the nominal in classic perceptual uses of complex demonstratives
plays a policing role, but that these expressions are either quantificational, or are
discourse anaphors in dynamic semantics. This would be an interesting combination of
features for these expressions to have: they would not be devices of direct reference, or of
reference of the sort that Evans (1982) and McDowell (1984) take demonstratives to
exemplify.48[48] Yet they would retain the main marks of referring expressions: they
would be rigid; they would involve a rapport between speaker and object; and they would
be object-dependent, in that if nothing is appropriated by the speaker’s use of the
expression, then the utterance in which it occurs will lack a truth value. Of course, it is
not news that expressions can be rigid, or sometimes require special rapport between
speaker and thing spoken about, even if the expression is not a device of reference. What
may be more surprising is that an expression (or some uses thereof) can be objectdependent, without being a device of reference, and without including any such device as
45[45]
That generalized quantifiers can carry presuppositions has been well-known for
some time, certainly since the work of Barwise and Cooper (1981).
46[46]
Nominal policing is compatible with the view (call it nominal stalking) that for an
utterance of That F is G to be true with respect to a world the appropriated object has to
be F in that world—and not merely in the context of utterance. (So if nominal policing is
true and nominal stalking false, then with respect to a world in which Frida is not by the
garbage, an utterance of (45) would be true, so long as in that world, Frida, wherever she
is, is hungry.) Braun (1994) and Borg (2000) defend nominal policing and deny nominal
stalking. Richard (1993) endorses nominal stalking. We argued against it in note 38.
King’s account of rigid uses of complex demonstratives denies nominal stalking as well.
47[47]
For instance, Roberts’ (2002) dynamic treatment of complex demonstratives makes
them a subspecies of definite noun phrases, all of which are treated as discourse anaphors
in a dynamic framework.
48[48]
The way Evans and McDowell think of such expressions, they have de re Fregean
senses.
a part. (A direct reference theorist would predict object-dependence for utterances of
Something identical with that, but would not expect a quantifier phrase to be able to be
object-dependent when it did not include any directly referential expressions.) The
compatibility of our account of nominal policing with a wide range of different semantics
for what many consider to be paradigmatic referring expressions brings this surprising
fact into focus.
We conclude on a speculative note. Referring expressions are typically taken to
have three features essentially: rigidity, object-dependence, and structural simplicity.
Armed with this assumption about what it is to be a referring expression, philosophers
have typically taken the distinction between referring and quantificational expressions to
be exclusive. 49[49] Our speculative suggestion is that the interesting semantic category in
the vicinity may rather be one that overlaps with quantificational expressions, rather than
one that excludes them. If complex demonstratives turn out to be quantificational in the
way that King has proposed, then given our arguments for nominal policing, there will be
quantificational expressions that have the first two of these properties (rigidity and
object-dependence).50[50] They would not count as referring expressions on the typical
characterization of such expressions, but that would only be due to their structural
complexity. More generally, if there turn out to be rigid and object-dependent
quantificational expressions, then structural simplicity alone would have to bear the entire
weight of making the distinction between quantificational and referential expressions
exclusive. Why think simplicity could bear so much weight? In particular, if structural
simplicity turns out to be something like syntactic simplicity at the level of logical form,
According to Stephen Neale’s (1993) “dilemma hypothesis,” every singular term is
either a quantificational expression or a referring one, where referring expressions are
taken to be unstructured, and the categories are exclusive. King (2001b) agrees that the
distinction is exclusive, but questions whether it is exhaustive. There has been a surge of
recent discussion of Neale’s hypothesis in connection with complex demonstratives: Borg
(2000) and Dever (2001) both argue that the case of complex demonstratives seems to
make trouble for this hypothesis, on the grounds that these expressions have semantically
significant structure (for Borg, this is a consequence of nominal policing). Neither rejects
the dilemma hypothesis in the end, though for different reasons. (Borg concludes that the
notion of what a referring expression is should be expanded, while Dever concludes that
complex demonstratives have a great deal of underlying syntactic and semantic
structure.) If our speculative hypothesis turns out to be correct, then while complex
demonstratives pose a problem for the dilemma hypothesis (as Borg and Dever suggest),
the moral to be drawn may be rather than referring expressions do not constitute a
semantic category distinct from quantificational expressions.
50[50]
The quantificational expression Every friend of Frida may be object dependent
thanks to the name Frida, but if so it is not object dependent in the relevant way. Its
object dependence simply derives from the object dependence of the name Frida and the
compositional semantics of the phrase. The object dependence with we are concerned, in
contrast, is not simply due to a sub-constituent being object dependent and contributing
its value to compositional semantics. So, for instance, on King's theory of complex
demonstratives, modified as propose to institute policing, we find object dependence but
not because of the object dependence of any sub-constituent.
49[49]
why think it is adequate to decide whether a phrase belongs to one semantic category or
another? We speculate that it could not. Whether or not the speculation is correct, the
case of complex demonstratives suggests that it is worth investigating.
References
Atlas J. D. and S. C. Levinson, 1981. It-clefts, informativeness, and logical form: Radical
pragmatics (revised standard version). In P. Cole (ed.), Radical Pragmatics. New
York: Academic Press.
Austin, J. L., 1975. How to Do Things with Words, 2nd ed. Cambridge: Harvard
University Press.
Barwise J. and R. Cooper, 1981. Generalized quantifiers and natural language.
Linguistics and Philosophy 4: 159–219.
Bever, D. I., 1997. Presupposition. In J. van Benthem and A. ter Meulen (eds.),
Handbook of Logic and Language, pp. 939–1008. Amsterdam: Elsevier.
Borg, E., 2000. Complex demonstratives. Philosophical Studies 97: 229–249.
Braun, D., 1993. Empty names. Nous 27: 449–469.
———, 1994. Structured characters and complex demonstratives. Philosophical Studies
74: 193–219.
Chierchia, G. and S. McConnell-Ginet, 2000. Meaning and Grammar, 2nd ed.
Cambridge: MIT Press.
Davies, M., 1982. Individuation and the semantics of complex demonstratives. Journal
of Philosophical Logic 11: 287–311.
Davis, S. (ed.), 1991. Pragmatics. Oxford: Oxford University Press.
Delin, J., 1992. Properties of it-cleft presupposition. Journal of Semantics 9: 289–306.
Dever, J., 2001. Complex demonstratives. Linguistics and Philosophy 23:271–330.
Donnellan, K., 1966. Reference and definite descriptions. Philosophical Review 75:
281–304. Reprinted in Davis (1991).
Dummett, M., 1959. Truth. Proceedings of the Aristotelian Society 59: 141–162.
Reprinted in Dummett (1978).
———, 1978. Truth and Other Enigmas. Cambridge: Harvard University Press.
Evans, G., 1982. The Varieties of Reference. Oxford: Oxford University Press.
Glanzberg, M., forthcoming a. Against truth-value gaps. Forthcoming in J. C. Beall
(ed.), Liars and Heaps: New Essays on Paradox. Oxford: Oxford University
Press.
———, forthcoming b. Presuppositions, truth values, and expressing propositions.
Forthcoming in G. Preyer and G. Peter (eds.), Contextualism in Philosophy.
Oxford: Oxford University Press.
Grice, H. P., 1975. Logic and Conversation. In P. Cole and J. Morgan (eds.), Speech Acts,
vol. 3 of Syntax and Semantics. New York: Academic Press. Reprinted in Grice
(1989).
———, 1989. Studies in the Ways of Words. Cambridge: Harvard University Press.
Heim, I., 1983. On the projection problem for presupposition. In M. Barlow, D.
Flickinger, and M. Westcoat (eds.), Proceedings of the West Coast Conference on
Formal Linguistics 2, pp. 114–125. Reprinted in Davis (1991).
———, 1990. E-type pronouns and donkey anaphora. Linguistics and Philosophy 13:
137–177.
Horn, L. R., 1989. A Natural History of Negation. Chicago: University of Chicago
Press.
Kadmon, N., 2001. Formal Pragmatics. Oxford: Blackwell.
Kaplan, D., 1977/1989. Demonstratives. In J. Almog, J. Perry, and H. Wettstein (eds.),
Themes from Kaplan, pp. 481–563. Oxford: Oxford University Press. This is the
first publication of a widely circulated manuscript dated 1977.
Karttunen, L., 1974. Presupposition and Linguistic Context. Theoretical Linguistics 1:
181–194. Reprinted in Davis (1991).
Kennedy, C. 1997. Projecting the Adjective: The Syntax and Semantics of Gradability
and Comparison. PhD Thesis, University of California, Santa Cruz. Published by
Garland Press, New York 1999.
King, J. C., 1999. Are ‘that’ phrases devices of direct reference? Nous 33: 155–182.
———, 2001a. Complex Demonstratives. Cambridge: MIT Press.
———, 2001b. Remarks on the syntax and semantics of day designators. Philosophical
Perspectives 15: 291–333.
Larson, R. and G. Segal, 1995. Knowledge of Meaning. Cambridge: MIT Press.
Lepore, E. and K. Ludwig, 2000. The semantics and pragmatics of complex
demonstratives. Mind 109: 199–240.
Levinson, S. C., 1983. Pragmatics. Cambridge: Cambridge University Press.
———, 2000. Presumptive Meanings. Cambridge, MIT Press.
Lewis, D., 1979. Scorekeeping in a language game. Journal of Philosophical Logic 8:
339-359. Reprinted in Davis (1991).
McDowell, J., 1984. De re senses. Philosophical Quarterly 34: 283–294. Reprinted in
McDowell (1998).
———, 1998. Meaning, Knowledge, and Reality. Cambridge: Harvard University Press.
Neale, S., 1993. Term Limits. Philosophical Perspectives 7: 89–123.
Richard, M., 1993. Articulated terms. Philosophical Perspectives 7: 207–230.
Roberts, C., 2002. Demonstratives as definites. In K. van Deemter and R. Kibble (eds.),
Information Sharing, pp. 89–136. Stanford: CSLI Publications.
Schiffer, S., 1981. Indexicals and the Theory of Reference. Synthese 49: 43-100.
Siegel, S., 2002. The role of perception in demonstrative reference. Philosophers’
Imprint 2 (1).
Simons, M., 2003. Presupposition and accommodation: Understanding the Stalnakerian
picture. Philosophical Studies 112: 251–278.
Soames, S., 1989. Presupposition. In D. Gabbay and F. Guenthner (eds.), Handbook of
Philosophical Logic, pp. 553–616. Dordrecht: Kluwer.
Stainton, R., 1995. Non-sentential assertions and semantic ellipsis. Linguistics and
Philosophy 18: 281–296.
Stalnaker, R. C., 1974. Pragmatic Presuppositions. In M. K. Munitz and P. K. Unger
(eds.), Semantics and Philosophy. Reprinted in Stalnaker (1999).
———, 1978. Assertion. In P. Cole (ed.), Pragmatics, vol. 9 of Syntax and Semantics.
Reprinted in Stalnaker (1999).
———, 1999. Context and Content. Oxford: Oxford University Press.
Strawson, P. F., 1950. On Referring. Mind 59: 320–344. Reprinted in Strawson (1971).
———, 1971. Logico-Linguistic Papers. London: Methuen.
Taylor, B. 1981. Truth-theory for indexical languages. In M. Platts (ed.), Reference,
Truth and Reality: Essays on the Philosophy of Language. London: Routledge
and Kegan Paul, 183-98.
van der Sandt, R. A., 1992. Presupposition projection as anaphora resolution. Journal of
Semantics 9: 333–377.